Evidence for Deleterious Original Antigenic Sin in SARS-Cov-2

bioRxiv preprint doi: https://doi.org/10.1101/2021.05.21.445201; this version posted May 24, 2021. The copyright holder for this preprint (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission.

1 Evidence for Deleterious Original Antigenic Sin in 2 SARS-CoV-2 Immune Response

3 Sanjana R. Sen1, Emily C. Sanders2, Alicia M. Santos2, Keertna Bhuvan2, Derek Y. Tang2, 4 Aidan A. Gelston2, Brian M. Miller2, Joni L. Ricks-Oddie3,4 and Gregory A. Weiss1,2,5* 5 6 Author Affiliations:

7 1 Department of Molecular Biology & Biochemistry, University of California, Irvine, Irvine CA 92697-3900 USA 8 2 Department of Chemistry, University of California, Irvine, Irvine CA 92697-2025 USA 9 3 Center for Statistical Consulting, Department of Statistics, University of California, Irvine, Irvine CA 92697-1250 10 USA 11 4 Biostatics, Epidemiology and Research Design Unit, Institute for Clinical and Translational Sciences, University of 12 California, Irvine, Irvine CA 92697-4094 USA 13 5 Department of Pharmaceutical Sciences, University of California, Irvine, Irvine CA 92697-3958 USA 14 15 * Address correspondence to [email protected] 16 17 Corresponding Author:

18 *GAW e-mail: [email protected], Tel: +1-949-824-5566, address: 1102 Natural Sciences 2, Irvine, CA 92697-2025, 19 USA 20 21 Author Contributions: 22 S.R.S., and G.A.W. designed research; S.R.S., E.C.S., A.M.S., K.B., D.Y.T, A.A.G., and B.M.M. performed research; 23 S.R.S., E.C.S. and G.A.W. analyzed data; J.L.R. advised on statistical analysis; S.R.S. and G.A.W. wrote the 24 manuscript. 25 26 Competing Interests: 27 None. 28 29 Acknowledgements: 30 We thank Professor Elizabeth Bess and Professor Stacey Schultz-Cherry for helpful 31 conversations, Kristin Gabriel for initial bioinformatics analysis, and the patients who generously 32 donated their plasma samples (IRB# 2012-8716). We gratefully acknowledge the support of the 33 UCI COVID-19 Basic, Translational and Clinical Research Fund (CRAFT), the Allergan 34 Foundation, and UCOP Emergency COVID-19 Research Seed Funding. G.S.S and A.M.S thank 35 the Minority Access to Research Careers (MARC) Program, funded by the NIH (GM-69337). 36 J.L.R. was supported by the National Center for Research Resources and the National Center for 37 Advancing Translational Sciences from the NIH (TR001414). 38

40 Abstract

41 A previous report demonstrated the strong association between the presence of

42 antibodies binding to an epitope region from SARS-CoV-2 nucleocapsid, termed Ep9, and

43 COVID-19 disease severity. Patients with anti-Ep9 antibodies (Abs) had hallmarks of original

44 antigenic sin (OAS), including early IgG upregulation and cytokine-associated injury. Thus, the

45 immunological memory of a previous infection was hypothesized to drive formation of suboptimal

46 anti-Ep9 Abs in severe COVID-19 infections. This study identifies a putative original antigen

47 capable of stimulating production of cross-reactive, anti-Ep9 Abs. From bioinformatics analysis,

48 21 potential “original” epitope regions were identified. Binding assays with patient blood samples

49 directly show cross-reactivity between Abs binding to Ep9 and only one homologous potential

50 antigen, a sequence derived from the neuraminidase protein of H3N2 Influenza A virus. This

51 cross-reactive binding affinity is highly virus strain specific and sensitive to even single amino acid

52 changes in epitope sequence. The neuraminidase protein is not present in the influenza vaccine,

53 and the anti-Ep9 Abs likely resulted from the widespread influenza infection in 2014. Therefore,

54 OAS from a previous infection could underlie some cases of COVID-19 disease severity and

55 explain the diversity observed in disease outcomes.

65 Introduction

66 Original antigenic sin (OAS) occurs when the immune responses adapted for a primary

67 (or “original”) infection instead target a similar, but not identical, pathogen (Fierz & Walz, 2020).

68 Since B-cells undergo affinity maturation post-primary infection, affinity-matured Abs from

69 previous infections can sometimes outcompete naïve Abs with specificity for the new antigen

70 (Brown & Essigmann, 2021). The phenomenon has been observed for immune responses to

71 dengue fever, influenza, respiratory syncytial virus and human immunodeficiency virus (HIV)

72 (Park, Kim, Park, Lee, & Park, 2016; Welsh & Fujinami, 2007). OAS ideally accelerates pathogen

73 clearance by targeting highly conserved antigens; however, suboptimal targeting by non-

74 neutralizing Abs binding to the antigen can exacerbate disease conditions by either enhancing

75 viral infection or hyperactivating the innate immune response (Brown & Essigmann, 2021). The

76 wide range of outcomes observed in SARS-CoV-2 infected individuals, from asymptomatic to

77 fatal, has been hypothesized to result from a patient’s unique immunological memory (Fierz &

78 Walz, 2020; Kohler & Nara, 2020).

79 Reports of OAS in COVID-19 patient have largely focused on the phenomenon’s beneficial

80 or benign impacts on the immune response. For example, Abs from individuals naïve for SARS-

81 CoV-2 infection, but exposed to other common human coronaviruses (hCoV), could cross-react

82 with SARS-CoV-2 spike protein and neutralize SARS-CoV-2 pseudotypes (Ng et al., 2020).

83 Plasma collected before the pandemic had Abs cross-reactive with SARS-CoV-2 nucleocapsid

84 (NP) and spike proteins; these Abs did not confer protection against severe COVID-19 outcomes

85 (Anderson et al., 2020). Though higher rates of prior cytomegalovirus and herpes simplex virus 1

86 infections have been reported for hospitalized COVID-19 patients versus patients with milder

87 symptoms, no Ab signature linked to a molecular mechanism has been shown (Shrock et al.,

88 2020). Humoral immunity to hCoVs, NL63 and 229E, has been associated with more severe

89 COVID-19 disease (Focosi et al., 2021).

90 Recently, we reported the association of severe COVID-19 with the presence of Abs

91 binding specifically to a 21-mer peptide derived from SARS-CoV-2 NP, an epitope region termed

92 Ep9. The patients, described as αEp9(+), comprised about 27% of the sampled, SARS-CoV-2-

93 infected population (n = 186). The αEp9(+) patients (n = 34 in this report chosen for having

94 sufficient sample availability) had high, early levels of αN IgGs, typically within the first week,

95 compared to αEp9(-) patients; cytokine-related, immune hyperactivity also appeared in the

96 αEp9(+) individuals (Sen et al., 2021). These two observations suggest an OAS-based

97 mechanism for the disease severity observed in αEp9(+) patients. Here, we explore the epitope

98 homology landscape and αEp9 Ab cross-reactivity to potentially identify the original antigen

99 driving Ab-based immune response in αEp9(+) patients.

100 Results and Discussion

101 Assays measured levels of αEp9 IgGs and IgMs from αEp9(+) patients whose plasma was

102 collected at various times post-symptom onset (PSO). Consistent with the hallmarks of OAS

103 tracing a prior infection, αEp9 IgG levels appeared elevated as early as one day PSO in one

104 patient; similar IgG levels were observed in the patient population over >4 weeks (one-way

105 ANOVA, p = 0.321) (Figure 2A). Levels of αEp9 IgMs amongst patients at various times PSO

106 were also similar (one-way ANOVA, p = 0.613). The signals measured for αEp9 IgM levels were

107 significantly lower than the equivalent αEp9 IgG levels (t-test, p = 0.0181) (Figure 2B); this

108 difference could reflect lower IgM affinity, quantity, or both.

109 Searches for homologs to Ep9 in protein sequence and structural homology databases

110 suggested candidate primary antigens or potential OAS epitopes. The Basic Local Alignment

111 Sequence Tool (pBLAST) (Altschul et al., 1997) and Vector Alignment Search Tool (VAST)

112 (Madej et al., 2014) identified structurally homologous regions from betaherpesvirus 6A and 14

113 other proteins with Ep9 sequence homology spanning >7 residues. Additionally, Ep9-orthologous

114 regions from six common human coronaviruses (SARS-CoV, MERS, OC43, HKU-1, NL63, 229E)

115 were also chosen for production and assays (Figure 1 A, Figure 1- supplement 1). To expedite

116 the binding measurements, the potential OAS epitope regions were subcloned into phagemids

117 encoding the sequences as fusions to the M13 bacteriophage P8 coat protein. DNA sequencing

118 and ELISA experiments demonstrated successful cloning and consistent phage display,

119 respectively. Two of the 21 epitopes failed to display on phage and were omitted from subsequent

120 investigations (Figure 1- supplement 3A).

121 The potential OAS epitopes to αEp9 Abs were tested by phage ELISA. To assess average

122 response within a patient population, pooled plasma from three sets of five αEp9(+) and five

123 αEp9(-) COVID-19 patients were coated onto ELISA plates, and tested for binding to the phage-

124 displayed potential OAS epitopes. Commercially sourced plasma from healthy individuals

125 provided an additional negative control. Confirming previously reported results, SARS-COV-2

126 Ep9 along with a homologous epitope from SARS-CoV-1 (90% similarity) bound only to plasma

127 from αEp9(+) patients (Sen et al., 2021). Since only 75 cases of SARS-CoV-1 were reported in

128 the US during the 2003 outbreak, this binding is unlikely to drive OAS (World Health Organization,

129 2013).

130 The panel of potential epitopes revealed a candidate epitope from the neuraminidase (NA)

131 protein of an H3N2 influenza A strain, which circulated in 2014 (A/Para/128982-IEC/2014,

132 Accession No. AIX95025.1). This epitope is termed EpNeu here. The plasma from αEp9(-)

133 patients and healthy individuals did not bind to EpNeu (p<0.0001, two-way ANOVA ad hoc Tukey

134 test) (Figure 1C, D). Surprisingly, SARS-CoV-2 Ep9 and EpNeu share only 38% amino acid

135 sequence similarity. Other candidate epitope regions had significantly higher homology, but failed

136 to bind to αEp9(+) plasma (Figure 1- supplement 1).

137

138 Figure 1

139

140 Potential OAS epitopes for binding αEp9 Abs suggested by bioinformatics and tested by 141 phage ELISA. (A) Cladogram depicting sequence homology of the Ep9 sequence from SARS- 142 CoV-2 to the bioinformatics-identified, closest homologs. Sequence alignments used pBLAST 143 and VAST, and the cladogram was generated by iTOL (Letunic & Bork, 2019). (B) Structures of 144 SARS-CoV-2 NP RNA binding domain (PDB: 6M3M) and the Flu A 2014 H3N2 NA protein 145 (modeled by SWISS-Model (Waterhouse et al., 2018)). SARS-CoV-2 NP highlights Ep9 residues

146 (light and dark blue) and the region homologous region to EpNeu (dark blue). The depicted model 147 of Flu A 2014 H3N2 NA highlights the EpNeu putative antigen (pink). (C) ELISAs examined 148 binding of phage-displayed potential OAS epitopes to Abs from three sets of pooled plasma from 149 five αEp9(+) patients, or five αEp9(-) patients. Pooled plasma from healthy patients was an 150 additional negative control. The colors of the heat map represent the mean binding signal 151 normalized to phage background negative controls (signal from phage without a displayed 152 peptide). (D) Expansion of data from panel C shows ELISA signals from the independently 153 assayed individual pools shows results from the individual pools (****p <0.0001 for a two-way 154 ANOVA comparing binding of phage-displayed epitopes listed in panel C to different groups of 155 pooled plasma, ad hoc Tukey test). (E) Using EpNeu as the search template to generate 156 homologous sequences (shown in next panel), ELISAs examined EpNeu homologs’ binding to 157 pooled plasma from αEp9(+), αEp9(-), or healthy individuals. The data are represented as 158 described in panel C (****p <0.0001 for two-way ANOVA c phage-displayed epitopes, ad hoc 159 Tukey and Dunnett’s test as shown). (F) Amino acid sequence alignment of the closely related 160 Flu A NA homologs of EpNeu from pBLAST(Altschul et al., 1997). Blue and orange residues 161 represent conserved and mismatched amino acids, respectively, relative to Ep9. Bolded residues 162 are important for epitope recognition by αEp9 Abs. Here, the term Flu refers to influenza. 163 164 The strain specificity of αEp9 Abs binding to H3N2 influenza A NA protein was tested next.

165 The EpNeu sequence provided the template for pBLAST homolog searching. Closely aligned NA

166 sequences isolated from human, avian, and swine hosts in North America were chosen for further

167 analysis (Figure 1F, Figure 1- supplement 1). The six sequences were phage-displayed as before.

168 Despite their close similarity to EpNeu (up to 92.3% similarity or only one residue difference),

169 none of the EpNeu homologs bound to Abs from αEp9(+) patients (Figure 1E). A single EpNeu

170 amino acid substitution, K142N (numbering from full-length NA, Accession No. AID57909.1)

171 occurs in an H1N2 swine flu from 2016; the single substitution dramatically decreased binding

172 affinity to Abs from αEp9(+) patients (p<0.0001 one-way ANOVA ad hoc Tukey). The epitope from

173 H4N6 avian influenza A from 2010 omits residue S141, but includes the conserved K142, also

174 greatly reduced binding to Abs from αEp9(+) patients (p<0.0001 one-way ANOVA ad hoc Tukey)

175 (Figure 1E, F). Therefore, S141 and K142 are critical for binding to αEp9 Abs.

176

177 Figure 2

178 179 Early upregulation of αEp9 IgGs and cross-reactive Ab binding to both Ep9 and EpNeu. 180 ELISA of αEp9 (A) IgG and (B) IgM levels in αEp9(+) patients (n = 34) from plasma collected at 181 the indicated time periods post-symptom onset (PSO). Statistical analysis was conducted using 182 one-way ANOVA, ad hoc Tukey test. The solid line represents the median value, and dashed 183 lines represent quartiles. (C) Comparing normalized levels of phage-displayed Ep9 and EpNeu 184 binding to plasma-coated wells from individual αEp9(+) patients (n = 34). A strong correlation is 185 observed, as shown by the depicted statistics. Each point in panels A through C represents data 186 from individual patients. (D) A schematic diagram of the sandwich ELISA to examine cross- 187 reactivity of αEp9 Abs. The assay tests for bivalent Ab binding to both Ep9 and EpNeu. Pooled 188 plasma from five αEp9(+) patients or five αEp9(-) patients with other αNP Abs was tested for 189 bivalent binding to both eGFP-fused Ep9 and phage-displayed EpNeu. Healthy patient plasma 190 was used as a negative control. For additional negative controls, phage-FLAG and eGFP-FLAG 191 replaced Ep9 and EpNeu, respectively (****p <0.0001 one-way ANOVA, ad hoc Tukey and 192 Dunnett’s test shown, with healthy plasma in the presence of EpNeu and Ep9 as negative control). 193 Error bars represent SD. Individual points on bar graph represent technical replicates. 194 195 Next, ELISAs were used to determine whether Ep9 and EpNeu epitopes bind to the same

196 Abs. Assays of 34 αEp9(+) patients demonstrated a strong and highly significant correlation

197 between levels of Abs binding to Ep9 and EpNeu epitopes in patient plasma, suggesting the two

198 epitopes may bind to the same Abs (Figure 2C). This cross-reactivity was confirmed by a

199 sandwich-format assay requiring bivalent, simultaneous binding by Abs to Ep9 and EpNeu (Figure

200 2D, Figure 2- supplement 1). In this assay, Ep9 was fused to the N-terminus of eGFP, and EpNeu

201 was phage-displayed. The results demonstrated binding of Abs to both Ep9 and EpNeu epitopes

202 in pooled plasma from αEp9(+) patients, but not in αEp9(-) patients with other αNP Abs or healthy

203 donors. Thus, we conclude that αEp9 Abs also recognize the EpNeu epitope region.

204 Figure 3

205

206

207 Binding interactions of αEp9 Abs with the predicted EpNeu epitope. (A) Linear and structural 208 B-cell epitope prediction tools Bepipred 2.0 and Discotope 2.0 suggested an extended, linear 209 epitope region from the influenza A H3N2 2014 NA, including the eight residues of Ep9 Neu (light 210 blue) with an additional ten, C-terminal residues (dark blue). This extended, predicted epitope is 211 termed EpPred. Structural epitope predictions are underlined. Residues on EpNeu that are not 212 aligned with Ep9 are depicted in orange. (B) Structural model depicting the influenza A H3N2 213 2014 NA. The model was generated using SWISS-Model based on the NA structure from 214 influenza A H3N2 Tanzania 2010 (PDB: 4GZS). The NA structure highlights the EpNeu region 215 (light blue), the extended residues in EpPred (dark blue), potential glycosylation sites (light pink), 216 and the residues S141 and K142 (red), which are important for αEp9 Ab recognition. (C) Dose- 217 dependent ELISA comparing binding of αEp9 Abs to Ep9, EpNeu and EpPred. Pooled plasma 218 from five αEp9(+) patients and five αEp9(-) patients were tested in triplicates with varying 219 concentrations of eGFP-fused epitopes. The data demonstrates the strongest interactions

220 occurred between αEp9 Abs and Ep9 with an approximately 2-fold decrease in αEp9 Abs binding 221 affinity for EpNeu. EpPred bound slightly stronger to αEp9 Abs than EpNeu; the difference in trend 222 lines of EpNeu and EpPred are statistically significant (p<0.0001, Comparison of Fits). Trendlines 223 represent non-linear regression fit with Hill slope analysis. 224 225 We then investigated whether EpNeu could present a viable antigen during the 2014 H3N2

226 (NCBI: txid1566483) infection. Linear epitope analysis of the whole NA protein using Bepipred

227 2.0 (Jespersen, Peters, Nielsen, & Marcatili, 2017) predicted a candidate antigen with eight

228 residues from EpNeu, including S141 and K142 plus ten additional residues (146-155). This

229 predicted epitope region, termed EpPred, includes the conserved catalytic NA residue D151

230 targeted for viral neutralization by the immune system (Gentles, Wan, Eichelberger, & Bloom,

231 2020) (Figures 3A, Figure 3- supplement 1A). Using a modelled structure of the 2014 H3N2 NA

232 from Swiss-Model (Bienert et al., 2017; Guex, Peitsch, & Schwede, 2009; Studer et al., 2021;

233 Waterhouse et al., 2018), the structural epitope prediction tool, Discotope 2.0 (Kringelum,

234 Lundegaard, Lund, & Nielsen, 2012), also identified potential epitopes within EpPred (Figures 3A,

235 Figure 3- supplement 1B).

236 EpPred was overexpressed as an eGFP-fusion protein (Figure 1-supplement 3B). Pooled

237 plasma from five αEp9(+) patients was tested for Ab binding to EpPred, EpNeu, Ep9, and a

238 negative control (eGFP FLAG). The αEp9 Abs bound to Ep9 with approximately two-fold higher

239 apparent affinity than to EpNeu (Figure 3C). The increased binding strength of Ep9 could result

240 from additional rounds of Ab affinity maturation post-primary infection (Brown & Essigmann,

241 2021). The longer length EpPred modestly improved upon the binding of EpNeu to αEp9 Abs

242 (Figure 3C). Thus, αEp9 Abs likely target a larger epitope of H3N2 2014 NA beyond regions

243 homologous to Ep9, though expression of NA makes this difficult to test (Lipničanová, Chmelová,

244 Godány, Ondrejovič, & Miertuš, 2020). Furthermore, the phage-displayed and bacterially

245 overexpressed epitopes applied here do not include post-translational modifications. Taken

246 together, the results support the hypothesis that the αEp9 Abs found in severe COVID-19 disease

247 may have originated from a primary infection with H3N2 influenza A.

248 Unfortunately, patient histories only rarely include previous influenza infections, and such

249 information was unavailable for patients in this study. The H3N2 2014 NA sequence was isolated

250 from Para, Brazil, and the strain’s spread in North America remains unknown. However, a severe

251 outbreak of influenza A H3N2 was recorded; in 2014, an antigenic shift from the vaccine strain

252 A/Texas/50/2012(H3N2) resulted in low vaccine efficiency (14%), and high levels of other strains

253 spread (Flannery et al., 2016; Xie et al., 2015). Since only hemagglutinin was sequenced for strain

254 identification in 2014 (Flannery et al., 2016), the candidate OAS strain from the current

255 investigation could not be effectively traced as only its NA sequence was available. Our data

256 reveals that the EpNeu homologous sequence of the vaccine H3N2 strain (identical to Flu A 2015

257 H3N2 NA, Accession No. ANM97445.1) cannot bind αEp Abs (Figure 1 E, F). Therefore, αEp Abs

258 must have been generated against a primary influenza infection and not the vaccine.

259 This report offers a molecular mechanism for OAS underlying the high-rate of severe

260 COVID-19 in αEp9(+) patients. Specifically, we demonstrate cross-reactive binding between

261 αEp9 Abs and a predicted NA epitope from a 2014 influenza A strain. Interestingly, a study of an

262 Italian patient cohort correlated COVID-19 disease outcomes with previous RSV and

263 herpesviruses, but not influenza A infections (Wang et al., 2020). Future studies could examine

264 correlation between a country’s rate of the H3N2 2014 influenza and severe COVID-19.

265 Additionally, health systems might have recorded such primary infections in specific patients,

266 which could also be searched for correlation. Examining epitope conservation and Ab cross-

267 reactivity could predict OAS-based immune responses and disease outcomes in future infections.

268 Identifying detrimental, benign or beneficial OAS pathways could also guide vaccine design for

269 increased efficacy and reduced risk.

270

271 Materials and Methods 272 273 Sequence and structural alignment analysis

274 To identify possible sources of primary infection responsible for Ep9 Ab generation,

275 sequence and structural alignment with Ep9 residues and the SARS-CoV-2 NP was conducted.

276 Alignment of Ep9 sequence with the orthologs from other human coronaviruses (hCoVs) such as

277 SARS-CoV, MERS, HKU-1, NL63, 229E and OC43 was conducted using the Benchling sequence

278 alignment tool (Benchling, 2017) (https://benchling.com). To explore a wider range of human host

279 pathogens pBLAST (Altschul et al., 1997) (https://blast.ncbi.nlm.nih.gov/Blast.cgi) was used to

280 search for Ep9 homology in a database of non-redundant protein sequences; common human-

281 host viruses were specified in the organism category. The queries were conducted with the blastp

282 (protein-protein BLAST) program (Altschul et al., 1997) with search parameters automatically

283 adjusted for short input sequences. Alignments spanning >7 residues were included here. The

284 Vector Alignment Search Tool (VAST) (Madej et al., 2014)

285 (https://structure.ncbi.nlm.nih.gov/Structure/VAST/vast.shtml) was used to find structural

286 alignment between SARS-CoV-2 Ep9 and proteins from other viral and bacterial human host

287 pathogens. Alignment for NP from common hCoV were not further examined, as they had been

288 included in sequence alignment analysis. The aligned sequences were sorted by the number of

289 aligned residues as well as root-mean square deviation (RMDS). The top 50 structurally aligned

290 proteins were then examined for structural homology in the Ep9 epitope region. Regions of

291 proteins that aligned with the Ep9 region were selected for subsequent analysis.

292

293 Cloning

294 Predicted OAS epitopes were subcloned for phage display using the pM1165a phagemid

295 vector (Levin, 2006) with an N-terminal FLAG-tag and a C-terminal P8 M13-bacteriophage coat

296 protein. OAS constructs were subcloned using the Q5 site-directed mutagenesis kit (New England

297 Biolabs, Ipswich, MA) as per manufacturer’s instructions. After cloning, cells were transformed 12

298 into XL-1 Blue E. coli and spread on carbenicillin-supplemented (50 μg/ml) plates. Individual

299 colonies were then inoculated into 5 ml cultures, and shaken overnight at 37 °C. The phagemid

300 was isolated using the QIAprep spin miniprep kit (Qiagen, Germantown, MD) as per

301 manufacturer’s instructions. Cloned sequences were verified by Sanger sequencing (Genewiz,

302 San Diego, CA).

303

304 Phage propagation and purification

305 The Ep9 homologs were expressed as N-terminal fusions to the P8 coat protein of M13

306 bacteriophage. Plasmids were transformed into SS320 E. coli and spread onto carbenicillin-

307 supplemented (50 μg/ml) LB-agar plates before overnight incubation at 37 °C. A single colony

308 was inoculated into a primary culture of 15 ml of 2YT supplemented with 50 μg/ml carbenicillin

309 and 2.5 μg/ml of tetracycline, and incubated at 37 °C with shaking at 225 rpm until an optical

310 density at 600 nm (OD600) of 0.5 to 0.7 was reached. 30 μM IPTG and M13KO7 helper phage at

311 an MOI 4.6 was added to the primary culture, and the culture was incubated for an additional 37

312 °C with shaking at 225 rpm for 45 min. 8 ml of the primary culture was then transferred to 300 ml

313 of 2YT supplemented with 50 μg/ml of carbenicillin and 20 μg/ml of kanamycin. The cultures were

314 inoculated at 30 °C with shaking at 225 rpm for around 19 h.

315 The phage propagation culture was centrifuged at 9632 x g for 10 min at 4 °C. The

316 supernatant, containing the phage, was transferred into a separate tubes pre-aliquoted with 1/5th

317 volume of phage precipitation buffer (20% w/v PEG-8000 and 2.5 M NaCl), and incubated on ice

318 for 30 min. The solution, containing precipitated phage, was centrifuged for 15 min at 4 °C, and

319 the supernatant was discarded. The precipitated phage was centrifuged a second time at 1,541

320 x g for 4 min at 4 °C, and then dissolved in 20 ml of resuspension buffer (10 mM phosphate, 137

321 mM NaCl, pH 7.4 - 8.0 with Tween-20 0.05%, v/v and glycerol 10% v/v). The resuspended pellet

322 solution was divided into 1 ml aliquots, which were flash frozen with liquid nitrogen for storage in

323 -80 °C. Prior to use in ELISA binding assays, the aliquoted phage-displayed constructs were re-

324 precipitated in 0.2 ml of phage precipitation buffer after incubation for 30 min on ice. Aliquots were

325 centrifuged at 12298 x g for 20 min at 4 °C and the supernatant was discarded. The phage pellets

326 were re-centrifuged at 1968 x g for 4 min at 4 °C, and then resuspended in 1 ml of 10 mM

327 phosphate, 137 mM NaCl, pH 7.4.

328

329 Expression and Purification of eGFP fusion peptides

330 pET28c plasmids encoding eGFP fusions to C-terminal Ep9-FLAG, EpNeu-FLAG,

331 EpPred-FLAG and FLAG (negative control) and N-terminal His6 peptide epitopes, were

332 transformed into BL21DE3 Star E. coli chemically competent cells. Transformants were spread

333 on carbenicillin-supplemented (50 g/ml) LB-agar plates and incubated at 37 °C overnight. Single

334 colonies of each construct were selected to inoculate 25 ml LB media supplemented with

335 carbenicillin (50 g/ml). After incubation at 37 °C with shaking at 255 rpm overnight, 5 ml of seed

336 cultures were used to inoculate 500 ml of LB media supplemented with carbenicillin (50 g/ml).

337 Expression cultures were incubated at 37 °C with shaking at 225 rpm until an OD600 of ~0.5 was

338 reached. The cultures were induced with 0.5 mM IPTG and incubated at 25 °C for 18 h. The cells

339 were pelleted by centrifugation at 9632 x g for 20 min and resuspended in Tris-HCl lysis buffer

340 (20 mM Tris-HCl, 250 mM NaCl, pH 8). Cells were lysed by sonication and the insoluble fractions

341 were pelleted by centrifugation at 24696 x g. The supernatant was affinity-purified using

342 Profinity™ IMAC (BioRad, Hercules, CA) resin charged with nickel sulfate. The protein lysate was

343 batch bound overnight to the IMAC resin and purified using gravity columns. Columns were

344 washed with lysis buffer supplemented with 20 mM imidazole, and the elution fractions were

345 collected from lysis buffer containing 250 mM imidazole. The elution fractions were then buffer-

346 exchanged with lysis buffer lacking imidazole using Vivaspin® 20 Ultrafiltration Units (Sartorius,

347 Goettingen, Germany) with a molecular weight cutoff of 10 kDa. The final buffer imidazole

348 concentrations were calculated to be ~0.1 mM. Purified and buffer-exchanged protein fractions

349 were then visualized using 10% SDS-PAGE with Coomassie dye staining.

350

351 Patient Sample Collection

352 Samples were collected as previously described (Sen et al., 2021). Briefly, the UC Irvine

353 Experimental Tissue Resource (ETR) operates under a blanket IRB protocol (UCI #2012-8716)

354 which enables sample collection in excess of requirements for clinical diagnosis, and allows

355 distribution to investigators. Plasma was collected from daily blood draws of COVID(+) patients,

356 initially confirmed with pharyngeal swabs. After immediate centrifugation, plasma from heparin-

357 anticoagulated blood was stored for 3-4 days at 4 °C prior to its release for research use. Personal

358 health information was omitted and unique de-identifier codes were assigned to patients to comply

359 with the Non-Human Subjects Determination exemption from the UCI IRB. At the research facility,

360 SARS-CoV-2 virus in plasma samples was inactivated through treatment by incubation in a 56 °C

361 water bath for 30 min (Pastorino, Touret, Gilles, de Lamballerie, & Charrel, 2020) prior to storage

362 at -80 °C.

363

364 Phage ELISAs

365 As described in previous reports (Sen et al., 2021), pooled plasma from five αEp9(+) patients, five

366 αEp9(-) patients, or healthy patients (Sigma-Aldrich, Saint Louis, MO) were separately prepared

367 in coating buffer (50 mM Na2CO3, pH 9.6); the plasma was diluted 100-fold during this step.

368 Plasma samples were then immobilized in 96 well microtiter plates by shaking the plasma

369 solutions at 150 rpm at room temperature (RT) for 30 min. After aspiration and washing by plate

370 washer (BioTek, Winooski, VT), each well was blocked with 100 μL of ChonBlock Blocking Buffer

371 (CBB) (Chondrex, Inc., Woodinville, WA) for 30 mins, shaking at 150 rpm at RT. Wells were

372 subsequently washed three times with PBS-T (0.05% v/v Tween-20 in PBS). Next, 1 nM phage-

373 displayed candidate “original” epitopes and controls prepared in CBB was incubated in microtiter

374 wells for 1 h at RT with shaking at 150 rpm. Unbound phage were aspirated and removed using

375 three washes with PBS-T. The peroxidase-conjugated detection antibody, αM13-HRP (Creative

376 Diagnostics, Shirley, NY), was diluted 1000-fold in Chonblock Secondary Antibody Dilution

377 (Chondrex, Inc., Woodinville, WA) buffer; 100 µl of this solution was added to each well before

378 incubation for 30 min at RT with shaking at 150 rpm. Following aspiration and three washes (100

379 µl each), 1-Step Ultra TMB-ELISA Substrate Solution (ThermoScientific, Carlsbad, CA) was

380 added (100 μl per well). Absorbance of TMB substrate was measured twice at 652 nm by UV-Vis

381 plate reader (BioTek Winooski, VT) after 5 and 15 min of incubation. The experiment was

382 repeated three times using plasma from different αEp9(+) and αEp9(-) patients for each

383 experiments, using a total of 15 patients for each group. Each experiment was conducted in

384 technical duplicate.

385

386 αEp9 IgG and IgM ELISA

387 Plasma from 34 patients, previously tested for the presence of αEp9 Abs using phage

388 ELISAs (Sen et al., 2021), were used to test levels of αEp9 IgGs and IgMs. 2 µM eGFP-Ep9 or

389 eGFP-FLAG in PBS pH 8.0 were immobilized onto 96 well microtiter plates via overnight

390 incubation with shaking at 150 rpm at 4 °C. Excess protein was aspirated and removed with three

391 consecutive PBS-T washes. Wells were blocked by adding CBB (100 µl) before incubation at 30

392 min at RT with shaking at 150 rpm. Next, αEp9(+) patient plasma, diluted 1:100 in CBB (100 µl),

393 was added to duplicate wells before incubation at RT for 1 h with shaking at 150 rpm. The

394 solutions were discarded and sample wells were washed with PBS-T three times. αEp9 Abs

395 binding to the potential epitopes was detected using horse radish peroxidase (HRP) conjugated

396 αHuman Fc IgG (Thermo Fisher Scientific, Waltham MA) or αIgM µ-chain specific (Millipore

397 Sigma, Temecula, CA) Abs diluted 1:5000 in ChonBlock Sample Antibody Dilution buffer. 100 µl

398 of detection Abs were added to each sample well, and incubated for 30 min at RT with shaking at

399 150 rpm. Sample wells were aspirated and washed three times in PBS-T, and the binding signal

400 was detected after addition of TMB substrate (100 µl per well).

401

402 Bivalent Abs binding ELISA

403 eGFP-Ep9 or eGFP-FLAG was serially diluted (120 nM, 40 nM, 13 nM and 4 nM) in PBS

404 pH 8.0, and added to the appropriate wells in 96 well microtiter plates, followed by shaking

405 overnight at 150 rpm at 4 °C. Excess unbound protein was removed, and the plate was washed

406 three times in PBS-T. Wells were then blocked in CBB and incubated for 30 min at RT. After

407 blocking, pooled plasma (100 µl per well) from either five αEp9(+) patients, or five non-αEp9,

408 αNP(+) patients, or healthy individuals was added to the appropriate wells. Plasma from pooled

409 patients was diluted 100-fold in CBB. As a positive control αFLAG Ab was used as a 1:2000

410 dilution in CBB. Samples were incubated for 1 h at RT with 150 rpm shaking. The solution was

411 removed by aspiration, and the plate and washed three times with PBS-T. Then 1 nM EpNeu

412 displaying phage or the phage negative control with no epitopes displayed was diluted in CBB.

413 100 µl phage solution was added to microtiter wells and incubated for 30 min at RT with shaking

414 at 150 rpm. After aspirating and washing off unbound phage, binding of phage-displayed EpNeu

415 to plasma αEp9 Abs was visualized using αM13-HRP Ab diluted 1:10,000 in ChonBlock Sample

416 Antibody Dilution buffer. Samples were incubated for 30 min at RT with 150 rpm shaking, and

417 unbound Abs were removed through washing with PBS-T three times before addition of TMB

418 substrate (100 µl). Experiments were conducted in technical triplicates and repeated three times

419 with different αEp(+) and αEp(-) patient samples.

420

421 Dose-dependent ELISA

422 Wells of microtiter plates were coated with serially diluted concentration of eGFP-Ep9,

423 EpNeu and EpPred or eGFP-FLAG, and incubated overnight at 4 °C before blocking as described

424 above. Next, pooled plasma (100 µl per well) from either five αEp9(+) patients, or five αEp9(-)

425 patients, or healthy individuals at 1:100 total plasma dilution in CBB was added to the appropriate

426 wells. Samples were incubated for 1 h at RT with shaking at 150 rpm. After incubation, unbound

427 solution was removed, and the plates were washed three times with PBS-T. αEp9 IgG levels were

428 detected by adding αFc IgG-HRP diluted 1:5000 in ChonBlock Sample Dilution buffer, followed

429 by incubation for 30 min at RT with shaking at 150 rpm, followed by addition of TMB substrate

430 (100 µl per well). Experiments were conducted in technical triplicates and repeated three times

431 with different αEp(+) and αEp(-) patient samples.

432

433 Linear B-cell Epitope Prediction

434 Linear epitopes from the Influenza A/Para/128982-IEC/2014(H3N2) neuraminidase

435 protein were predicted using the partial sequence with Accession AIX95025.1 from the National

436 Center for Biotechnology Information’s GenBank and the linear B-cell epitope prediction tool,

437 Bepipred 2.0 (Jespersen et al., 2017) (http://www.cbs.dtu.dk/services/BepiPred-2.0/). The

438 prediction thresholds were set to 0.5. The specificity and sensitivity of epitope prediction at this

439 threshold is 0.572 and 0.586, respectively.

440

441 Structure-based B-cell epitope prediction

442 The structure of Influenza A/Para/128982-IEC/2014(H3N2) neuraminidase protein was

443 modelled using Swiss-Model (Bertoni, Kiefer, Biasini, Bordoli, & Schwede, 2017; Bienert et al.,

444 2017; Guex et al., 2009; Studer et al., 2020, 2021; Waterhouse et al., 2018)

445 (https://swissmodel.expasy.org/interactive). Using the ProMod3 3.2.0 tool (Studer et al., 2021), a

446 structural model was generated based on the crystal structure (2.35Å, PDB 4GZS 1.A) of a

447 homologous H3N2 neuraminidase with 96.39% sequence identity. Modelling methods and quality

448 assessments are further detailed in the report below.

449 The structural model of Influenza A/Para/128982-IEC/2014(H3N2) neuraminidase was used to

450 predict structure-based epitopes. Using the in silico online platform Discotope 2.0 (Kringelum et

451 al., 2012) (http://www.cbs.dtu.dk/services/DiscoTope-2.0/), structure-based epitope propensity

452 scores were calculated to predict likely B-cell epitope residues. The score of -3.7 was set as the

453 threshold for epitope prediction, which estimates a specificity and sensitivity of 0.75 and 0.47,

454 respectively (Figure 3 – supplement 1).

455

456 Statistical Analysis

457 The ELISA data were analyzed in GraphPad Prism 9 (https://www.graphpad.com). Since

458 the ELISA assays of 21 potential OAS epitopes were conducted over several microtiter plates for

459 repeated experiments, the raw absorbance values for every patient sample were normalized and

460 represented as the ratio of phage negative control to the signal. For heatmaps, two-way Analysis

461 of variance (ANOVA) with a Tukey adjustment for multiple comparisons tests were conducted for

462 the entire dataset of epitopes. For column comparisons of two groups, for example IgM levels and

463 IgG levels in the αEp(+) patients, unpaired, two-tailed, parametric t-tests were applied.

464 Additionally, for column comparisons between more than two groups, for example IgM or IgG

465 levels groups by weeks PSO, One-way ANOVA with a Tukey adjustment for multiple comparisons

466 tests were used. Where indicated, an ANOVA with a Dunnett’s adjustment were performed to

467 compare results to healthy Abs interactions to αEp9(+) patient results. Graphs represent SD error

468 bars for technical replicates, defined as replicates of the same conditions in multiple wells of the

469 same plate. Whereas error bars are shown as SEM when an experiment is repeated with different

470 patient sample sets. Correlations between Ep9 and EpNeu levels in patients were determined by

471 plotting normalized values on an XY graph and performing a linear Pearson’s correlation

472 coefficient test, where a r coefficient between 1.0-0.7 were considered strong correlations, values

473 between 0.7 and 0.5 were considered a moderate correlation, and values below 0.5 were

474 considered a weak correlation (Mukaka, 2012). The significance of the correlation was evaluated

475 based on p-value <0.05.

476 Figure 1-supplement 1: Potential “original” epitopes targeted by αEp9 Abs

477

478

479

480 Figure 1 –supplement 2: Primers used to subclone potential original epitopes

481 Figure 1 – supplement 3

482

483 Expression of phage-displayed and eGFP-fused potential OAS epitopes. (A) ELISA 484 demonstrating the display of N-terminal FLAG-tagged potential epitopes fused to the N- 485 terminus of the P8 coat protein. Immobilized αFLAG Abs in microtiter wells bind the 486 displayed FLAG-tag and epitope, and binding is detected with αM-13-HRP Abs as usual. 487 Phage with no epitope displayed provide the negative control. Epitopes for 488 mastadenovirus protein (mAdV) P8 and V. bacterium NADH oxidoreductase (NOX) did 489 not display. Error bars represent SD values. (B) 10% SDS-PAGE gel stained with 490 Coomassie Blue shows His-tag affinity-purified and buffer-exchanged eGFP-fused 491 epitopes, EpPred, EpNeu, FLAG negative control and Ep9.

492

493

494

495

496

497

498 Figure 2 –supplement 1

499

500 Optimization of assay to determine cross-reactivity of αEp9 Ab to Ep9 and EpNeu. 501 Sandwich ELISA testing the binding of Abs from the pooled plasma of five αEp9(+) 502 patients, five αEp9(-) patients with other αNP Abs and healthy individuals. This 503 experiment examines bivalent binding to various doses of immobilized eGFP-fused Ep9 504 epitope (120, 40, 13 and 4 nM) and phage-displayed EpNeu in solution. The data shows 505 that Abs from αEp9(+) patients, but not αEp9(-) or healthy individuals, bivalently bind 506 both EpNeu and Ep9. The positive control (αFLAG 1:2000 fold dilution) at 100 nM eGFP 507 demonstrates concentrations appropriate for bivalent binding to immobilized and in- 508 solution tags. The schematic diagram illustrates the binding observed for bivalence in 509 αEp9 Abs, where the antibody bridges plate-bound eGFP at its high concentrations. 510 Therefore, Figure 2 in the main text uses 4 nM of eGFP Ep9 coated on the plate, and the 511 FLAG positive control uses eGFP at 100 nM. Error bars represent SD.

512

513

514

515

516

517 Figure 3 –supplement 1

518 519 Linear and structural epitope mapping prediction of Influenza A H3N2 520 Neuraminidase. (A) Linear epitope mapping prediction of the Flu A 2014 H3N2 using 521 Bepipred 2.0 (Jespersen et al., 2017) demonstrates high prediction scores in a region 522 spanning 18 residues, which includes eight residues from EpNeu (underlined). The 523 additional 10 predicted residues were included as part of an extended epitope termed 524 EpPred. (B) Structural epitope mapping, using Discotope 2.0 (Kringelum et al., 2012), of 525 the modelled neuraminidase protein from Flu A 2014 H3N2 (SWISS-model 526 (Waterhouse et al., 2018)), predicts an epitope of five residues. These were captured by 527 EpPred, including three found in EpNeu. 528

529 References 530 531 Altschul, S. F., Madden, T. L., Schäffer, A. A., Zhang, J., Zhang, Z., Miller, W., & Lipman, 532 D. J. (1997). Gapped BLAST and PSI-BLAST: A new generation of protein database 533 search programs. Nucleic Acids Research. https://doi.org/10.1093/nar/25.17.3389 534 Anderson, E. M., Goodwin, E. C., Verma, A., Arevalo, C. P., Bolton, M. J., Weirick, M. E., 535 … Vanderbeck, A. N. (2020, November). Seasonal human coronavirus antibodies 536 are boosted upon SARS-CoV-2 infection but not associated with protection. 537 MedRxiv. medRxiv. https://doi.org/10.1101/2020.11.06.20227215 538 Benchling, I. (2017). Benchling [Biology Software]. 539 Bertoni, M., Kiefer, F., Biasini, M., Bordoli, L., & Schwede, T. (2017). Modeling protein 540 quaternary structure of homo- and hetero-oligomers beyond binary interactions by 541 homology. Scientific Reports. https://doi.org/10.1038/s41598-017-09654-8 542 Bienert, S., Waterhouse, A., De Beer, T. A. P., Tauriello, G., Studer, G., Bordoli, L., & 543 Schwede, T. (2017). The SWISS-MODEL Repository-new features and functionality. 544 Nucleic Acids Research. https://doi.org/10.1093/nar/gkw1132 545 Brown, E. L., & Essigmann, H. T. (2021). Original Antigenic Sin: the Downside of 546 Immunological Memory and Implications for COVID-19. MSphere, 6(2). 547 https://doi.org/10.1128/mSphere.00056-21 548 Fierz, W., & Walz, B. (2020). Antibody Dependent Enhancement Due to Original Antigenic 549 Sin and the Development of SARS. Frontiers in Immunology, 11, 1120. 550 https://doi.org/10.3389/fimmu.2020.01120 551 Flannery, B., Zimmerman, R. K., Gubareva, L. V., Garten, R. J., Chung, J. R., Nowalk, M. 552 P., … Fry, A. M. (2016). Enhanced Genetic Characterization of Influenza A(H3N2) 553 Viruses and Vaccine Effectiveness by Genetic Group, 2014-2015. In Journal of 554 Infectious Diseases (Vol. 214, pp. 1010–1019). Oxford University Press. 555 https://doi.org/10.1093/infdis/jiw181 556 Focosi, D., Genoni, A., Lucenteforte, E., Tillati, S., Tamborini, A., Spezia, P. G., … Maggi, 557 F. (2021). Previous Humoral Immunity to the Endemic Seasonal Alphacoronaviruses 558 NL63 and 229E Is Associated with Worse Clinical Outcome in COVID-19 and 559 Suggests Original Antigenic Sin. Life. https://doi.org/10.3390/life11040298 560 Gentles, L. E., Wan, H., Eichelberger, M. C., & Bloom, J. D. (2020, May 9). Antibody 561 neutralization of an influenza virus that uses neuraminidase for receptor binding. 562 BioRxiv. bioRxiv. https://doi.org/10.1101/2020.05.08.084954 563 Guex, N., Peitsch, M. C., & Schwede, T. (2009). Automated comparative protein structure 564 modeling with SWISS-MODEL and Swiss-PdbViewer: A historical perspective. 565 Electrophoresis. https://doi.org/10.1002/elps.200900140 566 Jespersen, M. C., Peters, B., Nielsen, M., & Marcatili, P. (2017). BepiPred-2.0: Improving 567 sequence-based B-cell epitope prediction using conformational epitopes. Nucleic 568 Acids Research. https://doi.org/10.1093/nar/gkx346 569 Kohler, H., & Nara, P. (2020, August 1). A Novel Hypothesis for Original Antigenic Sin in 570 the Severe Disease of SARS-CoV-2 Infection. Monoclonal Antibodies in 571 Immunodiagnosis and Immunotherapy. Mary Ann Liebert Inc. 572 https://doi.org/10.1089/mab.2020.0029 573 Kringelum, J. V., Lundegaard, C., Lund, O., & Nielsen, M. (2012). Reliable B Cell Epitope 574 Predictions: Impacts of Method Development and Improved Benchmarking. PLoS

575 Computational Biology, 8(12). https://doi.org/10.1371/journal.pcbi.1002829 576 Letunic, I., & Bork, P. (2019). Interactive Tree of Life (iTOL) v4: Recent updates and new 577 developments. Nucleic Acids Research, 47(W1), W256–W259. 578 https://doi.org/10.1093/nar/gkz239 579 Levin, A. M. (2006). Exploring the interaction between the protein kinase A catalytic 580 subunit and caveolin-1 scaffolding domain with shotgun scanning, oligomer 581 complementation, NMR, and docking. Protein Science. 582 https://doi.org/10.1110/ps.051911706 583 Lipničanová, S., Chmelová, D., Godány, A., Ondrejovič, M., & Miertuš, S. (2020). 584 Purification of viral neuraminidase from inclusion bodies produced by recombinant 585 Escherichia coli. Journal of Biotechnology, 316, 27–34. 586 https://doi.org/10.1016/j.jbiotec.2020.04.005 587 Madej, T., Lanczycki, C. J., Zhang, D., Thiessen, P. A., Geer, R. C., Marchler-Bauer, A., 588 & Bryant, S. H. (2014). MMDB and VAST+: Tracking structural similarities between 589 macromolecular complexes. Nucleic Acids Research, 42(D1). 590 https://doi.org/10.1093/nar/gkt1208 591 Mukaka, M. M. (2012). Statistics corner: A guide to appropriate use of correlation 592 coefficient in medical research. Malawi Medical Journal. 593 Ng, K. W., Faulkner, N., Cornish, G. H., Rosa, A., Harvey, R., Hussain, S., … Kassiotis, 594 G. (2020). Preexisting and de novo humoral immunity to SARS-CoV-2 in humans. 595 Science, 370(6522), 1339–1343. https://doi.org/10.1126/science.abe1107 596 Park, M. S., Kim, J. Il, Park, S., Lee, I., & Park, M. S. (2016). Original antigenic sin 597 response to RNA viruses and antiviral immunity. Immune Network. Korean 598 Association of Immunologists. https://doi.org/10.4110/in.2016.16.5.261 599 Pastorino, B., Touret, F., Gilles, M., de Lamballerie, X., & Charrel, R. N. (2020). Heat 600 inactivation of different types of SARS-CoV-2 samples: What protocols for biosafety, 601 molecular detection and serological diagnostics? Viruses, 12(7), 735–743. 602 https://doi.org/10.3390/v12070735 603 Sen, S. R., Sanders, E. C., Gabriel, K. N., Miller, B. M., Isoda, H. M., Salcedo, G. S., … 604 Weiss, G. A. (2021). Predicting COVID-19 Severity with a Specific Nucleocapsid 605 Antibody plus Disease Risk Factor Score. MSphere, 6(2), e00203-21. 606 https://doi.org/10.1128/mSphere.00203-21 607 Shrock, E., Fujimura, E., Kula, T., Timms, R. T., Lee, I.-H., Leng, Y., … Elledge, S. J. 608 (2020). Viral epitope profiling of COVID-19 patients reveals cross-reactivity and 609 correlates of severity. Science, eabd4250. https://doi.org/10.1126/science.abd4250 610 Studer, G., Rempfer, C., Waterhouse, A. M., Gumienny, R., Haas, J., & Schwede, T. 611 (2020). QMEANDisCo—distance constraints applied on model quality estimation. 612 Bioinformatics. https://doi.org/10.1093/bioinformatics/btz828 613 Studer, G., Tauriello, G., Bienert, S., Biasini, M., Johner, N., & Schwede, T. (2021). 614 ProMod3 - A versatile homology modelling toolbox. PLoS Computational Biology. 615 https://doi.org/10.1371/JOURNAL.PCBI.1008667 616 Wang, L., Candia, J., Ma, L., Zhao, Y., Imberti, L., Sottini, A., … Zhang, Y. (2020). 617 Serological responses to human virome define clinical outcomes of italian patients 618 infected with SARS-CoV-2. MedRxiv. https://doi.org/10.1101/2020.09.04.20187088 619 Waterhouse, A., Bertoni, M., Bienert, S., Studer, G., Tauriello, G., Gumienny, R., … 620 Schwede, T. (2018). SWISS-MODEL: Homology modelling of protein structures and

621 complexes. Nucleic Acids Research. https://doi.org/10.1093/nar/gky427 622 Welsh, R. M., & Fujinami, R. S. (2007). Pathogenic epitopes, heterologous immunity and 623 vaccine design. Nature Reviews Microbiology, 5(7), 555–563. 624 https://doi.org/10.1038/nrmicro1709 625 World Health Organization. (2013). WHO | Cumulative Number of Reported Probable 626 Cases of Severe Acute Respiratory Syndrome (SARS). Retrieved May 16, 2021, 627 from http://www.who.int/csr/sars/country/en/ 628 Xie, H., Wan, X. F., Ye, Z., Plant, E. P., Zhao, Y., Xu, Y., … Zhu, Y. (2015). H3N2 629 Mismatch of 2014-15 Northern Hemisphere Influenza Vaccines and Head-to-head 630 Comparison between Human and Ferret Antisera derived Antigenic Maps. Scientific 631 Reports, 5(1), 1–10. https://doi.org/10.1038/srep15279 632